We’ve been experimenting with a new scheduling algorithm for the past week that allows for more simultaneous job execution. In general, the limits have been relaxed a bit, allowing for more jobs to be active in the queue on your account. While more jobs are allowed in simultaneous execution now, the jobs now share the Power setting of the highest Power job you have in the queue. Jobs with identical Power settings have that Power setting worth of compute resources scheduled with distribution skewed in order from the oldest to newest job in your queue. This prefers and contributes more compute resources to jobs that have been in the queue for a while, but allows for newer jobs to get at least some resources so you can get a few frames back quickly for spot-checking reasons. This new algorithm is most noticeable in the lowest of Power settings, as those were the most restricted by the old scheduling limits regarding simultaneous job execution. At higher Power settings, with multiple jobs in your queue, the impact of this algorithm is felt primarily in oldest jobs being finished before newer jobs due to resource skewing.
We have not updated our documentation regarding simultaneous job execution just yet, as we’re still plumbing the depths of how this algorithm works in the real world. We will update the docs eventually, once we have finalized how we want the scheduler to work.
We’ve also rolled out several updates to our C4D instances recently. R17 instances are now running the latest in the R17 build from Maxon. X-Particles was updated in R16 and R17. RealFlow was updated in all 3 C4D instances. Last, but not least, new and updated plugins from Greyscalegorilla have been deployed in all 3 C4D instances.